Forums

dac architecture

Started by John Larkin January 5, 2019
On Mon, 7 Jan 2019 21:25:49 +0100, Piotr Wyderski
<peter.pan@neverland.mil> wrote:

>John Larkin wrote: > >> We don't want a digital design (ADC, lookup table or polynomial, DAC) >> because that might add phase noise. > >Not sure if I understand correcty, but whenever there is a discrete >level change, there will inevitably be some injection of phase noise. >So what are you trying to avoid? > > Best regards, Piotr
DAC major bit glitches maybe. It's more interesting of the comparator gains are low. It was just a passing idea. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
John Larkin wrote:
>It would be tough to make a real 8-bit DAC out of resistors and CMOS >levels.
Why? Works perfectly, R2R is as old as the world: https://en.wikipedia.org/wiki/Resistor_ladder Got nice video from it (on output of Xilinx FPGA), driving straight into an emitter follower and then a 75 Ohm cable,
On Monday, 7 January 2019 20:16:05 UTC, John Larkin  wrote:

> It would be tough to make a real 8-bit DAC out of resistors and CMOS > levels.
Assuming you use 0.1% Rs, is CMOS R_out the problem? NT
John Larkin wrote:

> DAC major bit glitches maybe.
Now it makes sense, thanks! Best regards, Piotr
On 1/5/19 2:47 PM, John Larkin wrote:
> On Sat, 05 Jan 2019 19:12:25 GMT, <698839253X6D445TD@nospam.org> > wrote: > >> John Larkin wrote >>> We were talking about TCXOs. One measures temperature and drives a >>> varicap through some nonlinear transfer function to get minumum net >>> TC. >>> >>> We don't want a digital design (ADC, lookup table or polynomial, DAC) >>> because that might add phase noise. I guess you could use a static >>> polynomial with the equivalent of nonvolatile DPOTS as the >>> coefficients. >>> >>> >>> This occurred to me, not as anything practical maybe but as an >>> interesting architecture. >>> >>> https://www.dropbox.com/s/8ls632mndcxqby8/DAC_TCXO.JPG?raw=1 >>> >>> It's sort of a thermometer-code ADC, but each comparator incrementally >>> adds + or - one increment to the output. >>> >>> As the temperature increases, we jog the output up or down one >>> increment at a time. >>> >>> The sequence of switch settings become a delta-sigma code to make the >>> output. >>> >>> The comparators could be sort of linear, not step outputs, to kind of >>> interpolate a bit. Some flash ADCs did something like that, soft >>> comparators. >> >> Yes, >> but you can get more linear varicap effect by using for example 2. >> This paper shows some topologies and their effect: >> https://www.everythingrf.com/uploads/whitepapers/IEEE_BCTM_092010_2.pdf >> >> Then use a linear opamp feedback loop? >> >> I am using something like fig 1b on page 3 for my 25 MHz PLL reference for Eshail2. > > A TCXO wouldn't need a very linear varactor, but a tight PLL does.
Doesn't have to be super duper linear, though +-10% is lots tight enough, because all it does is change the loop gain a bit. You can do that well with an inductor in series and one in parallel.
> > I have a new circuit that starts a 600 MHz coaxial ceramic resonator > colpitts oscillator at trigger time, and phase locks it to an OCXO > asap, still preserving the phase of the oscillator relative to the > trigger. It uses an ADC to digitize the phase difference, an FPGA to > do the math, and a DAC+varicap to tweak the CCRO. It also uses a dual > varicap per fig 1b in your paper. Driving the varicap junction is > interesting. I designed the loop and can barely understand it myself. > > The TCXO thing I posted is interesting because it's delta-sigma in > space instead of the usual delta-sigma in time. > >
Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net https://hobbs-eo.com
John Larkin wrote:
> We were talking about TCXOs. One measures temperature and drives a > varicap through some nonlinear transfer function to get minumum net > TC. > > We don't want a digital design (ADC, lookup table or polynomial, DAC) > because that might add phase noise. I guess you could use a static > polynomial with the equivalent of nonvolatile DPOTS as the > coefficients. > > > This occurred to me, not as anything practical maybe but as an > interesting architecture. > > https://www.dropbox.com/s/8ls632mndcxqby8/DAC_TCXO.JPG?raw=1 > > It's sort of a thermometer-code ADC, but each comparator incrementally > adds + or - one increment to the output. > > As the temperature increases, we jog the output up or down one > increment at a time. > > The sequence of switch settings become a delta-sigma code to make the > output. > > The comparators could be sort of linear, not step outputs, to kind of > interpolate a bit. Some flash ADCs did something like that, soft > comparators. > > >
Just use a regular DAC and an LPF and set the knee above a frequency that matters. The output will mostly be zero, anyway. It's temperature so it's already heavily integrated. You just don't want too much process gain. -- Les Cargill
On 1/17/19 10:13 PM, Les Cargill wrote:
> John Larkin wrote: >> We were talking about TCXOs. One measures temperature and drives a >> varicap through some nonlinear transfer function to get minumum net >> TC. >> >> We don't want a digital design (ADC, lookup table or polynomial, DAC) >> because that might add phase noise. I guess you could use a static >> polynomial with the equivalent of nonvolatile DPOTS as the >> coefficients. >> >> >> This occurred to me, not as anything practical maybe but as an >> interesting architecture. >> >> https://www.dropbox.com/s/8ls632mndcxqby8/DAC_TCXO.JPG?raw=1 >> >> It's sort of a thermometer-code ADC, but each comparator incrementally >> adds + or - one increment to the output. >> >> As the temperature increases, we jog the output up or down one >> increment at a time. >> >> The sequence of switch settings become a delta-sigma code to make the >> output. >> >> The comparators could be sort of linear, not step outputs, to kind of >> interpolate a bit. Some flash ADCs did something like that, soft >> comparators. >> >> >> > > > Just use a regular DAC and an LPF and set the knee above a frequency > that matters. The output will mostly be zero, anyway. > > It's temperature so it's already heavily integrated. You just don't want > too much process gain.
You can't usefully lowpass filter 1/f noise. It's really a different regime, especially when you care about LF phase noise. Some years ago I was building stabilized lasers for geophysical applications (a downhole interferometric gravimeter). The basic idea is that you can measure the density of rock by measuring gravity at the surface (where the rock is pulling down) and then at depth, where some of the rock is now pulling up. It was also intended for reservoir management, where we'd leave one sensor at the bottom of the well and correlate its data with that at the surface. (There are important gravity variations due to barometric pressure, even.) The laser had to have an Allan variance below 10**-10 at 100000 seconds (about a day), so I locked a communications-type DFB laser to an air-spaced etalon made from optically-contacted Zerodur, which was itself temperature-controlled. (Optical contacting makes a hermetic seal, which gets rid of the drift due to air density.) The locking technique is one I invented almost 30 years ago: you sit halfway up an interference fringe, subtract the photocurrents from the transmitted and reflected beams, and servo at the null. That gets rid of the AM noise contribution. You have to attenuate the reflected beam a bit, because it's stronger than the transmitted beam due to cavity losses. As long as you're super paranoid about fringes due to unwanted surface reflections, it's a very very stable locking mechanism, and doesn't require super-high finesse etalons like Pound-Drever-Hall. Interestingly it turns out that if you adjust the attenuation so that dR/d omega + dT/d omega = 0 at the same frequency where R-T = 0 the out-of-band frequency noise decouples from the total amplitude measurement as well, so theoretically you can do intracavity measurements at the shot noise. (The loop suppresses the in-band noise.) Filtering was not a useful concept. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
Phil Hobbs wrote...
> > Some years ago I was building stabilized lasers for geophysical > applications (a downhole interferometric gravimeter). [snip]
Did you ever write that up? -- Thanks, - Win
On 1/18/19 9:31 AM, Winfield Hill wrote:
> Phil Hobbs wrote... >> >> Some years ago I was building stabilized lasers for geophysical >> applications (a downhole interferometric gravimeter). [snip] > > Did you ever write that up? > >
Nah, I haven't published a paper since I left IBM. The technology worked great, but the company went down the tubes (so to speak) when the founder and main technical guy went off on the most spectacular midlife crisis in my acquaintance--apparently he deserted his wife and five children, then skipped off to China and shacked up with a 22-year-old rich girl in Shanghai or someplace. (February 22nd is the 10th anniversary of my consulting business--we're going to throw a party.) Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
Phil Hobbs wrote:
> On 1/17/19 10:13 PM, Les Cargill wrote: >> John Larkin wrote: >>> We were talking about TCXOs. One measures temperature and drives a >>> varicap through some nonlinear transfer function to get minumum net >>> TC. >>> >>> We don't want a digital design (ADC, lookup table or polynomial, DAC) >>> because that might add phase noise. I guess you could use a static >>> polynomial with the equivalent of nonvolatile DPOTS as the >>> coefficients. >>> >>> >>> This occurred to me, not as anything practical maybe but as an >>> interesting architecture. >>> >>> https://www.dropbox.com/s/8ls632mndcxqby8/DAC_TCXO.JPG?raw=1 >>> >>> It's sort of a thermometer-code ADC, but each comparator incrementally >>> adds + or - one increment to the output. >>> >>> As the temperature increases, we jog the output up or down one >>> increment at a time. >>> >>> The sequence of switch settings become a delta-sigma code to make the >>> output. >>> >>> The comparators could be sort of linear, not step outputs, to kind of >>> interpolate a bit. Some flash ADCs did something like that, soft >>> comparators. >>> >>> >>> >> >> >> Just use a regular DAC and an LPF and set the knee above a frequency >> that matters. The output will mostly be zero, anyway. >> >> It's temperature so it's already heavily integrated. You just don't >> want too much process gain. > > You can't usefully lowpass filter 1/f noise.
Wait, what? Is that a thing in TCXO's ? I meant the lowpass really as just more integration - I do know you have to let these things get to equilibrium before you can trust them.
> It's really a different > regime, especially when you care about LF phase noise. >
Sounds like it then.
> Some years ago I was building stabilized lasers for geophysical > applications (a downhole interferometric gravimeter).&nbsp; The basic idea is > that you can measure the density of rock by measuring gravity at the > surface (where the rock is pulling down) and then at depth, where some > of the rock is now pulling up.&nbsp; It was also intended for reservoir > management, where we'd leave one sensor at the bottom of the well and > correlate its data with that at the surface.&nbsp; (There are important > gravity variations due to barometric pressure, even.) > > The laser had to have an Allan variance below 10**-10 at 100000 seconds > (about a day), so I locked a communications-type DFB laser to an > air-spaced etalon made from optically-contacted Zerodur, which was > itself temperature-controlled.&nbsp; (Optical contacting makes a hermetic > seal, which gets rid of the drift due to air density.) > > The locking technique is one I invented almost 30 years ago: you sit > halfway up an interference fringe, subtract the photocurrents from the > transmitted and reflected beams, and servo at the null.&nbsp; That gets rid > of the AM noise contribution.&nbsp; You have to attenuate the reflected beam > a bit, because it's stronger than the transmitted beam due to cavity > losses.&nbsp; As long as you're super paranoid about fringes due to unwanted > surface reflections, it's a very very stable locking mechanism, and > doesn't require super-high finesse etalons like Pound-Drever-Hall. > > Interestingly it turns out that if you adjust the attenuation so that > > dR/d omega + dT/d omega = 0 > > at the same frequency where > > R-T = 0 > > the out-of-band frequency noise decouples from the total amplitude > measurement as well, so theoretically you can do intracavity > measurements at the shot noise.&nbsp; (The loop suppresses the in-band noise.) >
That's crazy. Seems like it would also be a fine seismograph...
> Filtering was not a useful concept. >
I would rather think not :) I thought we were talking about a temperature controller.
> Cheers > > Phil Hobbs >
-- Les Cargill